Mastering ChatGPT: Advanced techniques for workplace communication and productivity | Hiten Shah

[00:00] I want to learn how to use ChatGPT the right way. So you gave us one tip, which was be thoughtful about its memories. But what else? What are your tips? I usually won't start anything without a ton of context or with the intention of giving it context over time. Just show it what great looks like. It's like a human. If a human doesn't know what great looks like, they're not going to know what great looks like. You do have an example of something that you think is great, which is your boss's operating manual. My boss's name is Morgan, so it's what would Morgan do? [00:26] So I'm gonna create the project and then I'm gonna add these files here. [00:29] I want to pitch Morgan the craziest [00:32] product idea. [00:34] I can think of. What is the best way to pitch it to him so we can go after it? [00:40] Now it's like, this is what you want Morgan to say. Yeah, that sounds like him. This is a tip for all the ICs out there. Go and replicate your boss and prep for conversations with them. [00:56] Welcome back to How I AI. I'm Claire Vo, product leader and AI obsessive here on a mission to help you build better with these new tools. Today, we have a great conversation with Hitton Shah, who's been building B2B SaaS for over 20 years. [01:10] Thank you. [01:10] Not only is he a great founder and product thinker, but he is, by my estimation, a total ChatGPT power user. [01:19] He's figured out how to load up ChatGPT projects with information about yourself [01:23] and about your boss to figure out the best way to communicate with each other and get work done. [01:28] Let's get to it.

[01:30] This episode is brought to you by Paragon, the integration infrastructure for AI SaaS companies. Are AI features on your 2025 product roadmap? Whether you need to ingest data for RAG from your users' external apps like Google Drive files, Gong transcripts, or JIRA tickets, [01:48] or build AI agents that automate work across your users' various tools, integrations are key to building useful AI features. However, building every one of these integrations costs months of engineering. [02:02] time you simply can't afford given the rapid pace of AI advancement. [02:07] Paragon is an all-in-one embedded integration platform for AI products. [02:12] Industry leaders like [02:14] AI21, you.com, and Copy AI use Paragon to connect over 100 of their users' SaaS applications. [02:22] to ingest data for RAG and provide their AI agents with thousands of integration actions. They've accelerated their integration development velocity by up to 50x, [02:33] allowing their engineering teams to focus on core product features. Want to fast track your integration roadmap this year? [02:41] Visit useparagon.com/howiai to learn how. [02:47] That's use Paragon, P-A-R-A-G-O-N, [02:53] dot com slash how I A I. [02:57] Thanks for being here. I am excited to talk about

[03:01] everybody's favorite friend, ChatGPT. Today is all about ChatGPT, right? [03:06] Yep. [03:06] How did you get to ChachiPT being your... [03:10] your favorite. You know, it's really funny. I [03:14] obviously started using ChatGPT first, [03:17] It existed first. There's a bunch of experiments I did with the API and all that prior to ChatGPT itself. So with OpenAI's APIs. And then when it came out, started playing around with it. [03:28] It really wasn't doing what I wanted in terms of my outputs that I was looking for. I think it was still helpful in a bunch of areas that we'll talk about today, but it wasn't quite getting there. So then I started using Claude when it came out. [03:39] And I was... [03:40] In the initial days of using this stuff, I was actually way more clod heavy. [03:45] by like [03:46] more than 50%. [03:49] Even though chat GPT came out first, I had access, I upgraded and all that. [03:53] And I think the turning point for me [03:55] which I can't prove because I don't know the dates, was when memory started working. [04:01] And that made it so that... And I'm a very like... [04:04] Um, [04:05] I don't try to talk to ChatGPT about everything. [04:08] And if I'm going to talk to it about things that are like a little random, I'll actually start a temporary session. I want to learn how to use ChatGPT the right way. So you gave us one tip, which was... [04:17] Be thoughtful about its memories. But what else? What are your tips? I'd love to see them. One of the biggest things here, we talked about memories, but temporary for things that you don't care for it to remember is really important. [04:29] And then I'll archive chats.

[04:32] because they don't count, so to speak, in the memories from last I checked. [04:36] You can also unarchive them. [04:38] At one point, I actually archived all my chats. [04:40] And then I started pulling out [04:44] The ones I liked. [04:46] it was definitely a pain for a number of reasons because [04:49] I have a lot of chats. [04:51] I have like thousands and thousands. That's why we're not going into me chat GPT today. I have a blank one with the $200 a month plan or whatever. [04:59] which was also a game changer. Once ChatGPT introduced that plan, [05:03] It definitely changed a lot of things for me in terms of my usage. I still get limits with Claude. [05:08] I barely get limits to ChatGPT, although I use projects a lot, so I definitely hit [05:12] decent amount of limits, but Claude's limits were also another very like [05:17] brutal thing for when you're trying to use it and trying to get something out of it, especially when you need it now. I think my next point is something everybody knows. [05:26] And that's that context is everything. [05:28] So for example, I don't have it shared here, but [05:31] I have a project [05:33] to make. [05:34] projects. [05:36] I have a project [05:37] to make deep research prompts? Well, this is some of the feedback that we've consistently gotten from guests, which is you can try to improve your output, [05:46] Or you can try to improve your input. And everybody says improve your input and the output will come. And so they really recommend, which I think you're going to show us, is how do you front load the time to make sure your prompts make sense, that your projects are set up accurately, that your GPTs are well structured. If you can do that, then you're going to get that consistent feedback.

[06:06] high quality outcomes. So what are those [06:09] kind of go to practices that you think everybody should do as they're building these things? Yeah, you know, if you don't love frameworks, then you need to start loving frameworks. I love frameworks. And I don't mean that in the, [06:22] Oh, Seeking Wisdom, Warren Buffett's Mental Models and Frameworks. Although I kind of do, but those are more like... [06:29] a bunch of tools. I think the strategy is if I'm trying to get something done, [06:33] Can I find something that's good enough or great? [06:36] that already works for other people and that someone's already shared. Oftentimes, [06:42] The models already know. [06:43] the framework. [06:44] So now you're just trying to like tease them into the right spot. [06:48] or [06:49] give them the framework itself. And then it sort of [06:52] gets even more contextual about it. And I've just noticed this over and over again. So I usually won't start anything. [06:57] without a ton of context, [06:59] or [07:01] with the intention of giving it context over time. [07:04] Because I don't want to say that like I won't just do a one sentence prompt. Yeah, I do that all the time, but I don't expect that one sentence prompt to just magically work. [07:12] I actually will even oftentimes be like, what context do you need? This is what my goal is. Or, and another one that like, I don't see enough people doing is [07:21] taking great outputs. [07:23] whether it's from AI or not from AI, and using that to help AI codify [07:27] that kind of output and it lets you build more outputs like that. [07:32] That's probably one of the magical ones that I found, which is just show it what great looks like. It's like a human. If a human doesn't know what great looks like,

[07:39] They don't have to know what great looks like. [07:40] So then you're just getting in some ways feedback hell, so to speak, of going back and forth until they get it or until you put something together. [07:47] But typically, as a human, if you give them an example and say, hey, [07:51] This is the thing I'd like from you. [07:54] they can do a pretty good job of filling in the blanks, so to speak. [07:57] I think AI can do a better job. So you do have an example of something that you think is great, which is your boss's operating manual. So could we use that as an example to get into exactly how you might do this? [08:10] Yeah, we'll create the whole thing. [08:13] I would first create a project. [08:15] And my boss's name is Morgan, so it's, what would Morgan do? [08:18] Or what would Morgan say? So it's WWMD. [08:21] So I'm gonna create the project, [08:23] And then I'm going to add these files here. [08:26] And so he has an operating manual. [08:29] Really great human being. [08:30] built this out, sent it over, [08:32] when he first joined the company, [08:35] and became my boss. And then I've been sending this to everyone. And now everyone's already seen it. And new people that join because someone else is sending it to them. Great. So that's his manual. And for anyone that doesn't know, these are things that oftentimes, you know, it used to be a bigger thing back in the day, I think, where people would create their operating manual. So you knew how to work with them. [08:53] Now there's some other tactics and stuff that are actually part of other projects that we can talk about. Then he also tends to share a few different [09:00] sort of [09:01] document, not documents, but like articles he likes. [09:04] And so there's one he shared called job is communication. I'm sure it's out there somewhere in the world. [09:08] If it's not, whatever. So I would put those two things in and that's the project files, right?

[09:13] Then I would add instructions. [09:15] So I'm going to open a new chat and be like, [09:17] I actually would do exactly this. I would add this, [09:20] Add these two in here. So this is a chat. [09:23] I'm going to stick to what I know, which is 4-0. 4-0. Yeah, yeah, yeah. [09:27] classic yeah yeah i'm a i'm very classic on that so i'll do that i'll pick 4-0 i really like 4-0 but all of them tend to work 4-0 is just faster and gets me what i want quicker [09:38] And so here I'll say, [09:41] I'm creating project. [09:43] with these documents and I want to be able to converse with it with the project and simulate type of feedback or advice that my boss would give me. [09:58] Can you create instructions for that project? [10:02] knowing that I'll be using these files in the project. [10:07] I love that you corrected your typo because I just roll right through it. That was very polite of you. [10:14] It's funny, like, [10:15] The way I type [10:16] is always correcting anyway. So it wasn't for ChatDVT's sake. Because, you know, yeah. So yeah, but that's great. OK, so here it already gives me the objective. Fine. Boss is Morgan. Great. [10:30] Uh... [10:31] System configuration personal. Okay. Okay. Interesting. [10:35] It has not given me the actual instructions. [10:39] Right? [10:39] So I'm like, okay, [10:41] where the actual instructions. So I'm going to go and this is like in another doc in there, which I call AI shaping.

[10:47] because you're basically trying to shape it to get at what you want. So this is a classic example of, I could paste all this crap in there, so to speak, [10:55] But, and it says project setup, but it doesn't feel right. So it's like, [10:58] Can you give me the specific instructions that I can use in the project? [11:04] So I can just paste those in there. [11:07] So I find this a useful kind of like prompt, which is just tell it where it's going to go. And maybe it'll give you a little bit better. Oh, no, we got some. Oh, well, there's some instructions. [11:19] Yeah, so usually what I've already done is, and this is another trick, I've done this in several chats. [11:26] So when I find one where it actually did what I wanted, which we can argue, we'll probably just try this one or I can copy one over if this doesn't work. [11:32] from where I have it. [11:33] But... [11:34] This is what I find most useful. If I can get a good output, [11:38] I will always try to get it to codify the output. [11:41] especially if it's one where I'm going to have to do this over and over again. [11:45] So in this case, I said, hey, give me the instructions. [11:48] I think just for shits and giggles, so to speak, [11:50] I'm going to use these because I think they might be good enough. And I'm not trying to review it. And I also don't really care too much about this as long as it looks good enough, because I can always iterate [12:02] and do things. And I also use the word simulate. I don't always use the word. In this case, I felt like, okay, let's try it like this. So there's just a lot of like, okay, we'll do it like that. [12:10] But if this were really, really good and I'm like, oh crap, this is impressive. I'd make it codify it and I'd turn it into, I could turn it into a project or something. So one thing I would do if I love this is, hey, can you help me create a project so I can create these type of instructions in the future?

[12:25] And that's what's got me a very good system of projects helping me [12:29] Use. [12:30] - Project for projects. - Project for projects, yeah. Whatever, I know it's so-- - Great, so you're gonna add the instructions right here. - Yep, right here. So they're all in here, cool. [12:37] Let's say, and then now I'm just going after it. I want to pitch... [12:41] On a pitch, Morgan, the craziest product idea I can think of. [12:47] have for us. What is the best way to [12:51] to pitch it to him so we can go after it. [12:56] Now it's like, this is what you want Morgan to say. Oh yeah. It's like, thank you. Okay. That sounds like him. Yep. And this part, this is key. [13:05] So he's our VP of product and growth. I report to him. [13:08] I'm a PM and [13:10] This is... [13:12] I know his manual. This is the type of stuff you would go for. So now it's like very specific lead with the pressure, AI first CRM, whatever, whatever crap, right? [13:21] So it basically gives me exactly [13:24] what I should do when it comes to him. [13:26] And it even says, what will he respect, et cetera. [13:29] Oh, man. So we've had a lot of people on the pod or a couple of people who have explained how as a manager, you can replicate yourself. But this is a tip for all the ICs out there. Go and replicate yourself. [13:43] your boss and prep for [13:46] conversations with them. And, you know, it's funny, I read this and I was thinking, oh, is this kind of generic? And I was like, some of that would work on me. Some of it wouldn't. But I bet people that work with me

[13:56] could. [13:57] pluck a couple artifacts and make a make a project or a GPT that really does. Oh, [14:06] Find out everything about me in a professional sense. [14:09] There you go. So you're doing it. [14:11] So what I would do if you were my boss is I would literally do this because I know there's enough about you out there. Now, in another kind of example we can give or it'll be in the doc. I do the same for myself. I call it like a personal OS. [14:23] and it has my personality type, any of the personality tests, [14:27] Myers-Briggs and everything. And then it helps me a lot. But if I know that about you, [14:31] So Morgan is actually an Enneagram five. I'm an Enneagram nine. [14:36] What I would actually do is feed that into the instructions in that as well. If I want to make it specific to me. [14:42] and our relationship. [14:43] Because [14:44] These personality tests help you, well, any environment in particular, [14:48] helps you with relationships and how to sort of, uh, [14:52] meet in the middle around communications and stuff like that. Yeah, that's actually a really great idea because I do all this when I start as a new leader on a team. I do the work of gathering, you know, has anybody in a disc or Enneagram or any, you know, Myers-Briggs. We kind of do that. We also have this questionnaire, which I'm happy to share in the show notes, which is 10 questions about me. What am I motivated by? [15:12] How do I like to receive feedback? How do I work all this stuff and actually do. [15:16] I have this repository in a Google Drive file of everything I know. [15:21] about the team that they've been told or that they've told us and I haven't yet put it in a GPT. And that's so smart because then I could say,

[15:29] you know and and sorry sorry zach and amanda i can say [15:33] You know, Zach and Amanda are not seeing eye to eye on Project X. [15:37] What would be an effective way, given what you know about them, [15:41] to come together. So it's a really interesting... [15:46] idea. Uh oh. [15:48] psychological warfare in the workplace, all for the greater good, empowered by OpenAI. [15:55] So we'll let it do its thing. What I'm doing is I'm doing research on you, and I said to get stuff in the professional sense or whatever. [16:04] Sometimes I'll even include, hey, [16:07] she's blah blah blah at this place that way it doesn't get the wrong person so it might but we'll see i don't think there's too many claire votes but maybe there are so we should be good uh and then yeah so we'll see what happens there we can continue with whatever else [16:21] Great. So I love, you know, you showed this about someone else and you're doing deep research on me, which. [16:27] you know, is always going to be thrilling to watch on the other side. But I want to go back to this thing that you said about your personal OS, the self-aware, you know, addition. [16:36] How did you build that? Why did you build that? [16:40] Who uses it? How are you adding to it? I'm so curious about this specific use case. [16:46] So I have a bunch of these links. This one is the personal OS. So I use my project creator to create this. [16:53] And it's basically a project description, what to use it for, and then these are the actual instructions. [16:59] Right? So what I would do is basically

[17:01] take this or take the instructions [17:04] here, [17:05] and put them into a project. So we'll just go do that. [17:09] so that we can show not tell, right? [17:11] And then it even says things like files to upload. [17:15] And then even has like a kickoff prompt if you want to play around or whatever. [17:18] So... [17:19] I will go here. [17:21] make a new project, [17:23] I don't need to put any files in there, but I'll probably do that too. But what you'll notice is up here, it's like, what are these things? [17:29] Right. So this is nine. This is one three. If you don't know human design, you should know human design. But I'm not going to get into that because it's on the woo woo side. This is the third time human design has come up. I live in San Francisco. I don't know if this is an SF meme that's happening right now has come up in the last. [17:45] week. So yes, it is quite woo woo. [17:48] They'll send you a PDF for free, but... [17:51] Why not? Yeah, I highly recommend. Okay, so I also have a voice and tone guide that was made for me a long time ago because I used to write a lot more. [18:01] And so I'll just put that in there just for shits and giggles, so to speak, although it probably isn't as relevant. So here we are with this. Let's play some games. [18:10] I'm going to take Morgan's manual, [18:12] and basically talk to it as if I just got this boss. [18:16] because this is kind of what I did. So this manual is from my boss, Morgan. [18:22] I'm newly reporting to him and would love any and all advice on how best to work with him, considering my own personality, etc.

[18:34] By the way, he's an Enneagram type. [18:39] So already it's like I didn't have to do the work and it's already giving me the contrast here. I would have to do a bunch of Google searches to be like type Enneagram type five and type nine relationship. [18:50] And then I got to wait, right? [18:51] But this is kind of what's interesting where it started mixing in the one, three, three, [18:56] from human design, which I love. And that's why I love giving it that context. [19:00] Then there's like, here's our working style alignments. [19:03] right all this is true like i've done this before i know i know that this is all very accurate [19:10] And then this is where things start getting interesting because it starts coming up with shit. Then that's why like the one sentence ones are great because it's like, wait, a check-in format. [19:18] Huh. [19:19] So this is how I could give him a weekly update on whatever, or this is how I could put our one-on-one doc and manage up. And he's going to love it because this is how he thinks. [19:28] Is this what you would like? Do you think it's accurate? Yeah, I've read his manual. Yeah, yeah, yeah. And I know it by heart. And this is like exactly it. And one of the big ones for him, and again, [19:39] Everyone will know this if... [19:41] They know how to lead, but that's a different story. [19:43] But... [19:44] If you're a good manager, [19:46] My opinion is that, and so is Morgan's, is that you're just unblocking. [19:50] All day. [19:51] And he said that in the manual. My job is to unblock you so we can get shit done. [19:55] Not in those words. Those are my words. So yeah, this is perfect. This is exactly what he'd want. [20:00] Because if there's a blocker, I know he's the first person I should hit up. If I can't solve it any other way, he will just solve it.

[20:06] And so that's cool. [20:08] There's some really interesting things around conversations. [20:11] He does ask you to defend things, but he never makes you feel... [20:15] Like you're offended when he asks you to defend things. So he's always got a good way of doing it. So that's an interesting point there. [20:21] This is kind of interesting. This is interesting, right? So this is in his doc. So it's quoting directly from the manual. Yeah, and that's why I make these projects. Got it. If I didn't make it a project, which I've done before and done the same thing, [20:35] I don't get this level of context coming back to me. [20:39] I get more of like, oh, you put this in. I'm going to go fill in the blanks. I'm going to do something like that. Here I get these kind of references and all that kind of stuff that I just [20:47] It's just much harder to get that in a chat. 'Cause I used to try this with a chat, [20:50] Maybe before they had projects. I know they had and cut some GPT is a whole nother mess. I don't like getting into. [20:56] But yeah. [20:57] So one of the things that I noticed here that I want to call out is you said you like frameworks. And so what I saw in some of this prompting is here are personality type frameworks or guides that have been well established or well documented that LLMs probably know about. [21:27] to, in a pretty structured way, address common workplace things, whether I'm going to have a hard conversation, I'm going to pitch something, I'm in conflict with someone. And I think that idea of mixing structured frameworks and personal context is really useful and

[21:43] One of the things that I'm really empathetic towards new leaders are I get lots of new leaders in my organization. They're in the biggest job of their career. And they always ask, can I get coaching? And I don't know if you... [21:54] The coaches out there are really... [21:56] You're really making money out there. Coaches are expensive. Good coaches are expensive. But a lot of times they're doing this work. They're doing 360s to gather information about you. They're applying frameworks to help you understand yourself and someone else. [22:08] And then they're helping you get out of mental patterns or giving you other ways to look at problems. [22:16] in a structured way that allows you to move forward. And I think you just basically built like [22:21] both kind of [22:22] coach for you to work with your boss but i can imagine just a coach for yourself so if you loaded this up with your own personality so this works like this is a coach for me right like [22:31] Like, I'm really mad right now about someone. Let's come up with a scenario. Let me make one up. Someone... [22:38] trying to get control over one of the projects I'm working on. [22:45] Because they are trying to lick the cookie. [22:50] as I have been told, happens. [22:55] So I just did that without even giving it a bigger problem. [22:58] And then this is where it gets interesting. [23:00] This reaction makes perfect sense. Okay, fine, chat GPT, you want to be positive, cool. But then it tells me why. [23:08] It threatens my control. [23:10] Not just my control, my integrity. [23:12] of what I'm building. And this is accurate. Like when someone says, oh, I'm starting your control. I'm like, yeah, I'm a control freak, but that's not the problem here. But then it said the integrity. There you go.

[23:21] okay so you've done deep research on me i'm going to give you my enneagram and my myers-brigg and because we're fun in my team [23:29] I'm a Sagittarius. We're going to toss that into the mix. And I just want to test. I want to test if this passes my... [23:35] Brand near to it. Sniff. [23:37] Okay. Well, while we're waiting for that to happen, let's switch over to one more use case, which is your sales use case, which I think is super interesting. Okay. [23:45] Yeah, this one's really fun. I find myself over time getting really excited about [23:52] frameworks that I have applied and work and also that the people who built them applied them and I can prove it. [23:58] So whether that's lead startup from Eric Reese or [24:02] I guess they call it the Shawn Ellis score now. It's it's it was always called the product macrofit score in my world, but that and a bunch of these other frameworks. So when I landed on this winning by design framework, I was like over the moon. [24:13] And so I did what a lot of people would do if you were really curious about it. So I did a site colon, their website. [24:20] file type PDF in Google. [24:22] And this is before deep research and all that, of course. [24:24] So I found all their PDFs. [24:26] which was before AI too. So I was like, okay, great. Now I can use these and do what I'm about to show you. [24:33] It took hours to do this kind of stuff. And now I can do it in seconds, literally. So... [24:37] Basically, [24:39] I have this entire kind of context put in. So this one's already set up. I can obviously set it up again, but [24:46] The deal is there's a bunch of these [24:47] documents. [24:48] And these documents are all about the framework and the various processes. These are all publicly available.

[24:53] They also have a bunch of private ones you can get if you go through their courses. [24:56] Then, [24:57] I use my... [24:59] instructions project, you got it. [25:02] to basically build this. It's pretty simple and straightforward. It's not really crazy because a lot of the context is in the files. [25:08] So, and again, this is easy to tweak anytime. [25:12] This will be familiar to anybody that's in a revenue organization. All leaders come in and they have their favorite [25:18] sales framework. [25:20] enable the field on it, you get a bunch of PDFs, but then I'm hoping you're going to show us how to apply this consistently over time. Okay, great. So I actually did this already. [25:30] So I actually asked you, what can you help me with? [25:33] So I didn't even want to like, again, I do a lot of that sometimes with these projects, just because I want it to tell me what it thinks it can help me with. [25:41] Just like a human, right? Like, what can you do for me? Or what's your best skill? Whatever. So it's like I can create and refine, create or refine sales processes. [25:48] wow okay so i can build a discovery call script cool [25:51] Perfect discovery call playbook, draft that, cool. Design customer-centric stories messaging. Run impact meetings, coach your GTM team. Cool, cool, okay, great. Where do I start? Okay, can you create a spice discovery guide? [26:02] And I did it for your product. [26:05] and it didn't know much about your product. [26:07] Because I didn't go do a web search. [26:09] I solved that problem too, as you probably can guess. But, and then it's basically like, okay. [26:14] VP of product, director of product management. Cool. This is a problem space. So now I know that if I give it this context myself, it will probably tweak things. So already,

[26:24] I think I'm winning because [26:26] I have now learned something of how to use this thing, which is another piece I go into. [26:30] And then now it goes into situation because the whole [26:33] premise of this framework is there are questions you should be asking on sales call. So you're only demoing like five or 10 minutes of a 20 or 30 minute call, not even five or 10, five minutes. [26:42] So what you're trying to do is diagnose pain. So we imagine these are like those discovery calls. [26:46] So these are a set of calls. [26:47] Or a set of questions, which is like how many PMs are on your team, whatever. Great. [26:51] I literally, every time I do this, I'm surprised by at least one or two questions in there, which would have been super... [26:59] challenging for me to think of on my own. Oh, so that's where I find a lot of the value. Otherwise, you can look at me like, Yeah, this is obvious. Yeah, a lot of it is, [27:05] But that's not the point. The point is that if you're going to go get on a call, [27:08] and you want to use the sales process, you want to do a really good job, hopefully. [27:12] These questions help you figure out how to do that. Yeah, this question under pain, I don't ask, but I should, which is where do things typically break down? I'm not good at pain discovery. So this is already giving me some ideas. [27:24] Yeah, exactly. And so like in pain is like, you know, as you know, one of the most important things when you're trying to actually do sales. [27:31] And then the impact questions are always valuable. But again, this is part of the value of the framework. [27:35] So I think the key thing to highlight is [27:37] I love that you like that. It makes perfect sense. [27:40] It's the framework producing it. It's not ChatGPT. I mean, it is. [27:44] But it's that framework that taught ChatGPT how to answer these questions very precisely, because it has like 20 PDS. [27:51] from these people. [27:52] You know? [27:54] And then a critical event is a big one. So if you're in sales, you know that you need to figure out if there is a date

[27:59] tied to the initiative and there's an initiative tied to it. So this is great. [28:03] So yeah, I had to do it, right? [28:05] So, [28:06] Then, to your point, oh, maybe it doesn't have context. Maybe it doesn't know enough. Great. [28:10] You can imagine what I did. I did a deep research prompt. I asked for a deep research prompt on it. I did it in the sales winning by design project on purpose. Even though the project is not meant for that, of course it can do whatever the heck [28:22] I try to get it to do. [28:24] But I purposely did that because I wanted to improve the winning by design outputs. [28:29] Thank you. [28:29] So I already did this yesterday. It's right here. [28:32] So I did some kind of analysis. [28:34] got me all kinds of things about your product company, whatever's going on. Same prompt it gave me, did all the corrections. So now all I do, and this is how dummy I am about it, [28:45] because it usually works. I'll just copy it. You hit the copy button. Don't even need to read it. I don't care. It's all good. Don't worry about that. Okay. [28:52] Now, [28:53] Here is context on chat PRD in terms of the history, etc. [28:59] of the product and [29:02] research [29:03] related to it. Fine. Whatever it could say. Whatever. Then [29:08] I would probably go at the end and say, [29:10] Please improve. [29:12] the deliverable above. [29:15] using this new content, this updated context? I am a dash, carrot, dash, you know, arrow, little... [29:23] instructor as well. All the time. This arrow. It just feels like it splits it up, but it probably doesn't matter, but I just want to do that. I love it. I agree.

[29:33] So this is my point about the memory, right? It'll tell you when it does that. [29:36] So oftentimes I'll sit there and just start pruning it. And so that's what I like about it, is it just makes it really easy to go cut that out of the memory if I want to. I don't need that. Yeah. Right, exactly. [29:45] So here you go, purpose, spice journey map. It did a little bit of a different thing, which is fine. But at the end of the day, [29:52] It's even telling me [29:54] what the gap analysis is. Maybe it got confused about something, but I don't know. Maybe not. [29:59] But yeah, that's the way I would do this stuff. And I'd probably prompt it a little differently and be like, okay, and I'll probably say this. Can you give me a demo script for the most relevant ICP slash persona? Whatever, in case it doesn't get it. [30:14] that the research identified. And I don't even know if the research identified this, but chat GPT test-- You just clicked the copy button. [30:23] Yes. [30:25] uh savvy sales and you're oh interesting [30:27] Well, so I think what it did, if I read above, [30:30] You didn't prompt it very well. No, not at all. Which is it tried to integrate the framework into my product. Yeah. As opposed to integrate the framework into the pit. [30:40] So let's do it again. [30:42] Yeah, but this is one of the things I love about AI is [30:46] you know imagine i told some research analyst go do this for me and they came back three days later and i was like no no sweet child totally wrong so expensive [30:56] You get this wrong, whatever. [30:59] 90 seconds we just lost and we can just give it another go. Yep, exactly.

[31:06] Thank you. [31:06] And I think this goes back to what you said at the beginning, which was [31:10] Who cares if you have to reprompt it? [31:12] It's so cheap. It's so fast. Like, who cares if we have to do this again? [31:16] It totally just works. [31:19] Yeah. [31:20] And I'll keep doing that. I don't care. It takes nothing, right? [31:23] And I think that's great. I kind of love the mistakes it makes, actually. Sometimes I learn all kinds of weird shit from the mistakes it makes. [31:30] Yeah, it keeps saying RevOps and product persona, which is interesting. [31:34] Well, I think this gives us whether or not to give us a good output. It gives us some good ideas. Let's see if our research, your deep research on Claire is done. I don't think there's actually that much. [31:45] interesting about me on the internet. But let's see. Well, we're about to find out. [31:50] Yeah. [31:51] Okay. [31:51] Oh, we got it. Okay. So you got a novel. [31:55] That was the point. - On Claire. I do have board roles, so. - Oh, you do? Okay. - I do. Look at me. - Okay. Okay. Okay, I just wanted to check. - Just go walk. [32:04] Yeah, no, I'm never surprised at how much it gives me these days because it tends to give me a lot when I do these kind of prompts. I do these kind of a lot. Okay, so you're copying it. [32:13] What is the best way we should do this? Should we make a project? Yeah, make a project. Okay. [32:19] Claire has. Claire. It's just Claire. Yeah. Okay. Download that. [32:26] Let me put that in here. [32:29] Okay. [32:30] Very exciting. Yeah. Yeah. Okay. [32:33] Clara is my boss. [32:35] There is research on her from the internet.

[32:40] in the project files. Want your help as a farm working [32:45] with her. [32:47] - Okay. - Okay. - Enneagram one. [32:49] Claire is at [32:52] Penny and Dry and Wine, okay. [32:54] I'm a ENTP Myers-Briggs. ENTP, okay. Hard D disk. Hard D disk, okay. [33:03] And a Sagittarius, of course. [33:05] fire sign sagittary ah that's what ius yeah whatever it'll be carrying something like that [33:15] all right yeah there you go i don't know any sagittarius obviously wait yeah okay we got her might as well capitalize it okay [33:22] Cool. [33:24] What do you want to ask it? [33:25] I have to tell Claire no on a roadmap item she wants. [33:29] Like I don't even know if this will work. [33:31] -Well, we're going to find out. -Yeah. No, it's because it's just so dirty with the research, right? And then, but we did get the Enneagram, so. [33:38] Okay, so now you're going to have to judge this. Okay, I'm going to judge it. All right. Let's see if we think that she's known to empower strong convictions. They're backed by evidence and framed in terms of business fact. Okay. [33:48] I want to be up front. I do like I want to be up front. [33:52] That's great. [33:54] I famously, though, this is very meta for this conversation, famously hate frameworks. So that framework, prioritization framework is not going to work for me. [34:02] Okay, does it meet the bar? Yep, that's good. [34:04] If helpful, I'd like to set it up as a structured experiment. Totally would work on me. Lightweight experiments, that's the easiest way.

[34:11] to get what I want. [34:13] Yeah, use experimentation language. [34:15] It opens the door without giving false hope. [34:18] Sure, Hope Springs Eternal. [34:21] This is... I don't know, you know? It doesn't... [34:26] It's decent advice. [34:27] it would probably work on me. [34:29] That's kind of my point, right? So now here's another little fun one I did. I actually did this. I did deep research on myself and I happen to have it sitting there. [34:36] And I happened to be at dinner with a very old friend of mine from college. [34:40] And I did something like this. I don't remember the exact prompt. And I basically took my deep research and said, [34:46] Please describe Hinton. [34:48] - Oh, goodness. - Right? - Yeah. - And then I just showed him, 'cause I wanted to show him the power of this thing, [34:55] He's like, that's you. [34:56] Like literally everything it said was me. [35:00] I am pragmatic. Okay. Leadership style. High velocity. That's the one I want. When people ask what it's like to work with me, [35:08] I say I go very fast. So that really... [35:12] Got me. Okay, let's go. Product philosophy. Let's see if this one gets right. [35:18] Mmm. [35:19] Cross-functional by design? Yeah, I think that's right. [35:23] Yeah. Okay. Let's go to my core traits. [35:26] Oh, they got me right. [35:28] Builder at heart. [35:30] That's the one I want! [35:33] The quote is ridiculous. We'll highlight this on the podcast. Be more... [35:39] entitled. I have found this to work for pretty much everyone I've done, as long as it's the right person. Yeah. Okay, this was great. Okay, so you've shown us, I just am going through everything you've shown us. You've shown us how to make...

[35:53] projects to make projects or prompts to make prompts, [35:57] You showed us how to manage knowledge. [36:00] and prune knowledge, as you say, which I think is a really useful tactic. You've given us a lot of great GPT frameworks for... [36:07] knowing thyself, knowing your manager, knowing each other and using that to give feedback to each other and then [36:14] You've shown us that despite my personal anti-framework bias, this has actually compelled me to think maybe I'm just applying frameworks the wrong way. And if I had an easy tool where somebody else did it for me, I would use it. So you've shown how you can scale sort of like training and frameworks into always on frameworks. [36:31] GPTs or models. So this is [36:34] It's been all about ChatGPT, but it's been super useful. We're going to do a couple lightning round questions and then I'm going to get you out of here. So question number one, you said that most of your AI usage is ChatGPT. You mentioned [36:50] Claude a little bit. What else is in your stack? - I use AI three to six hours a day is what I'm clocking in at. It's always three at the minimum. [36:58] It used to be about two years. [36:59] And [37:01] The specific tools are ChatGPT and Claude. [37:07] The most useful tool I use is not Publix. I can't talk about it or the name. [37:11] It's actually a desktop app. [37:13] um, [37:14] And that's outside of chat GPT. [37:17] Uh... [37:19] Outside of that, [37:21] I try probably 10 to 20 new tools a week.

[37:25] I guess here's what I'll say. Yep. [37:28] I think [37:30] The tools are fine. [37:31] But when I go back to AI tools, [37:35] Nothing beats a blank canvas for me. [37:38] from chat GPT today. And there's a bunch of reasons why, but the one [37:42] that I keep coming back to that I see people making a big mistake of, which we kind of did here to demonstrate it, which is, [37:49] I [37:50] So many people try to go to automation. [37:53] When I need to understand what's the prompting that's going to work. [37:56] And how do I get this thing to spit out reliable prompts? Again, we call it evals and stuff in the product world, which I have lots of opinions on. [38:03] But, [38:03] the bare metal, the chat, messing with that. Like right now in five minutes, I could fix that sales winning by design thing. So it actually gives the output we want is probably the instructions. [38:12] It probably doesn't understand that when I give it context, it needs to use the context and nothing else. [38:17] Right? Simple thing. I could fix it. [38:20] If I tried to automate that, [38:22] and say I want to do automated sales scripts where I just toss it in and do it and use one of these automated tools. I know there's all these fancy awesome ones, even Zapier, etc. [38:29] I'm probably going to be stuck with all the outputs suck. [38:32] Okay, and then I'm done. [38:34] But I don't want to be done. I want to go build the plan on how to build the automation. And the only way to build the plan is to manually do it repeatedly. [38:43] Then there's the witch model and all that, which is a whole nother can of worms. [38:46] Because you use API for obvious reasons with the product you're building. [38:51] You have a specific model that you know every time you throw the prompts at it that you like to throw at it, it gives you the output that the customers like.

[38:58] So, [39:00] i think people are erasing the automation and that's a mistake yeah and i'll just call out my friend pete kuman at yc did this article basically begging [39:07] begging AI product people to let users [39:12] manage the problems a little bit better. So yeah, I have so many opinions on that. [39:16] And... [39:18] Yes and no. I think the big one that I would point people to if they need actual, true, good inspiration on that is a product called Lex. Lex.Page.L-E-X. [39:27] They've instrumented the whole product so you can tear up all the prompting, make it your own, etc. in a very impressive way. So... [39:35] The yes and no is like, [39:36] Some products, it makes sense. I think others... [39:39] it will lead you in a bad direction. And I haven't told Pete this yet, but I've read his whole thing. [39:43] And I have similar thoughts in general. Some of the products we're building, the Lex model makes sense. And I call it the Lex model because Nathan's one of the best product people out there. Don't tell him I said that, although... [39:53] He's gonna hear it. - We love him and we love Black. - Yeah, so, but the way he thinks about this is very much aligned with what Pete was saying. I would just advise people that, [40:03] Some products [40:04] don't make sense to make the user see all that or do all that kind of work. So then the onus is back on us product people. [40:10] All right. And then speaking of work, I am going to wrap with my final question, which is when ChadGPT is not working. [40:19] for you yes you're a prompt master but if you need to get inside i won't ask you what enneagram chat gpt is but if you need to get inside its head [40:27] What's your prompting technique?

[40:29] Do you get mad? Do you do all caps? [40:32] to offer money, [40:33] Yeah, I know people like to treat it like a human. [40:36] And that's because the interface is making us treat it like a human, not because it is a human. [40:42] And I think there's a whole world out there to talk about that. That would take us many, many hours and a bunch of drinks. [40:50] But high level [40:52] I... [40:54] bluntly, [40:56] Just say, this is incorrect. So right now, I would say, hey, this is incorrect. Your instructions must be wrong. Because I'll give it an assumption of mine, of what's wrong. [41:05] please blah blah blah or do blah blah blah. But I don't like try to bribe it or any of that. I know those tricks work, but [41:11] To me, [41:12] that feels artificial. [41:14] And I'd instead like to find the same... [41:18] The same way to get what I want without the bribe, just like with a human, I wouldn't want to bribe them all the time because they're going to get trained on wanting the bribe. And I know ChatGPT. [41:26] with the memories and all that, it's going to get trained on your bribing. So please don't bribe it unless you want to head down that path, just like with a human. And so I guess the TLDR is... [41:35] If you're going to treat it like a human, think about incentives. [41:37] You are this just made me think of my eight year old and my six year old. And you're right. The second they know a bribe works, they look for the bribe. OK, this has been so great. Very useful. [41:48] Where can we find you and how can we be helpful? [41:51] I'm on Twitter. [41:52] H.N. Shah, H-N-S-H-A-H, I tweet way too much. [41:56] Yeah, that's it. Like, whatever else, like, just a person that's been in tech probably way too long and loves playing with these things. But more importantly, I think we're, we're,

[42:07] Everyone says we're in an incredible time and this and that, but it's like, [42:10] If you have access to the tool, you should use the tool. And if you need help using the tool, like hit me up. I don't know. Most of my friends hit me up and I built things for them, but like. [42:19] That's why, you know, [42:20] Thank you for having me on here. I love it. Well, thank you so much. Thank you, Claire. Thanks so much for watching. If you enjoyed this show, please like and subscribe here on YouTube, or even better, leave us a comment with your thoughts. [42:33] You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app. Please consider leaving us a rating and review, which will help others find the show. You can see all our episodes and learn more about the show at howiaipod.com. See you next time.